MEASUREMENT UNCERTAINTY
as a bottom-up approach due to the GUM method. However, this has changed recently, and the MU and TE approaches can both be thought of as top-down. There are, however, multiple top-down methods for MU estimation that are impacted differently when considering bias. In a study of 20 common chemistry measurands some met maximum uncertainty specifications using ISOTS 20914:2019 and not using the Nordtest method, and some showed the reverse.4
the impact of uRW
u(bias) in the Nordtest method, showing that different uncertainty methods also result in different influences and interpretation.
The biggest challenge is, in a mathematical sense, how should bias be incorporated into an overarching quality specification. Bias is linear. It has magnitude and direction. Imprecision is not linear. It is a variance. Increasing variance results in widening of the probability distribution in both directions, unlike bias that, if increasing, moves further away from the ‘truth’. From a mathematical perspective there are reasons that bias and imprecision cannot be combined directly, either linearly as in the case of total error, or in quadrature as with MU. For MU, if absolute bias is linearly added to the imprecision (uRW
) it
would overestimate MU significantly, and this is reinforced if we believe that short- term biases are included within long-term imprecision. We potentially double count small biases. Bias and imprecision are also not independent; they affect each other as their values change. At the moment there is no satisfactory resolution to the discussions.
Conclusions This article has approached bias both in the context of a clear difference from a true value but also in the context of small-scale biases that are incorporated into random variability over time. This part of measurement uncertainty quantification by the top-down method is far from the finished article and many discussions continue about the validity of approaches. The discussions on bias, its uncertainty and its correction raises more philosophical questions than it answers at the moment. How long does something have to be present for it to be a bias that needs correcting? At what point does it become clinically significant? How long can it be left in the hope that it swings the other way and is absorbed by uRW
.
The only thing that is clear at the moment is that bias is rarely handled correctly, irrespective of on which side of the fence you sit.
18
Analyser A n = 120
Mean = 100.3 SD = 5.1
Variance = 26.2
Analyser B n = 80
Mean = 108.2 SD = 5.2
Variance = 27.0
Analyser C n = 95
Mean = 98.6 SD = 4.8
Variance = 23.0
Pooled average variance = 25.4
Differences are attributed to in the ISO method and
Variance of means = 17.5
Pooled uncertainty
across analysers = 6.6
Combined with ucal
for combined uncertainty of platform
Fig 3. Schematic representation of applying a single combined uncertainty that can be applied to a patient sample, irrespective of what analyser it is measured on, incorporating pooled uncertainties across analysers and accounting for different running means of IQC. n = number of IQC runs, SD = Standard Deviation.
References 1 Ercan Ş. Comparison of Sigma metrics
computed by three bias estimation approaches for 33 chemistry and 26 immunoassay analytes. Adv Lab Med. 2023 Jul 4;4(3):236–45. doi: 10.1515/ almed-2022-0095.
2 Gidske G, Sandberg S, Fauskanger P et al. Aggregated data from the same laboratories participating in two glucose external quality assessment schemes show that commutability and transfers of values to control materials are decisive for the biases found. Clin Chem Lab Medicine 2023 Jul 21;62(1):77–84. doi: 10.1515/ cclm-2023-0532.
3 Coskun A. Bias in Laboratory Medicine: The Dark Side of the Moon. Ann Lab Med. 2024 Jan 1;44(1):6–20. doi: 10.3343/ alm.2024.44.1.6.
4 Nurlu N, Cat A, Ucar KT. Measurement uncertainty in clinical chemistry: ISO 20914 versus nordtest or intermediate precision versus bias. Scand J Clin Lab Invest. 2024 May 14:1–7. doi: 10.1080/00365513.2024.2338738. Epub ahead of print.
Further reading n Krouwer Consulting. A simple improvement
to total error and measurement JUNE 2024
WWW.PATHOLOGYINPRACTICE.COM
uncertainty (https://jkrouwer.wordpress. com/2018/01/15/a-simple-improvement- to-total-error-and-measurement- uncertainty)./
n Krouwer JS. The problem with total error models in establishing performance specifications and a simple remedy. Clin Chem Lab Med. 2016 Aug 1; 54(8): 1299–301. doi: 10.1515/cclm-2015-1175.
n Lim CY, Markus C, Greaves R, Loh TP; IFCC Working Group on Method Evaluation Protocols. Difference- and regression- based approaches for detection of bias. Clin Biochem. 2023 Apr; 114:86–94. doi: 10.1016/
j.clinbiochem.2023.02.007.
n Miller WG, Budd J, Greenberg N et al. IFCC Working Group Recommendations for Correction of Bias Caused by Noncommutability of a Certified Reference Material Used in the Calibration Hierarchy of an End-User Measurement Procedure. Clin Chem. 2020 Jun 1; 66(6):769–78. doi: 10.1093/clinchem/hvaa048.
Dr Stephen MacDonald is Principal Clinical Scientist, The Specialist Haemostasis Unit, Cambridge University Hospitals NHS Foundation Trust, Cambridge Biomedical Campus, Hills Road, Cambridge CB2 0QQ. +44 (0)1223 216746.
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56
